The Evolution of Voluntary Cooperation in Firms: Evidence From a Field Experiment in a Large Retail Company

How to cite

Brahm, Francisco; Riquelme, Cristina; Loch, Christoph. 2017. "The Evolution of Voluntary Cooperation in Firms: Evidence From a Field Experiment in a Large Retail Company." AEA RCT Registry. AEARCTR-0002350. https://doi.org/10.1257/rct.2350-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Francisco Brahm* , Cristina Riquelme , Christoph Loch

* Primary investigator

Registry ID
AEARCTR-0002350
DOI
https://doi.org/10.1257/rct.2350-1.0
First registered
Last updated
Status
On going
Trial dates
2017-04-01 – 2019-03-30
Countries
Private
JEL codes
C93, D23, D90, H41, I10, L20, M14, Z13
Registry record
http://www.socialscienceregistry.org/trials/2350

Abstract

Voluntary worker to worker helping behavior -i.e., worker cooperation- is very important for the performance of any organization. There is a vast theoretical literature in the fields of evolutionary biology and evolutionary anthropology regarding the conditions and mechanisms that favor the evolution of cooperation in populations. However, extant empirical evidence comes mainly from the lab, with scant evidence coming from the field.
In this project, we collaborate with three organizations in order to experimentally intervene the implementation of a workplace safety methodology that is based on voluntary cooperation by workers. In this prevention methodology, a starting group of ten workers is voluntarily created which then is trained to provide structured feedback on safety behavior to fellow workers in the site (in our case, the sites have approximately 250 workers). Then, this group strives to expand within the site, inviting and training workers to become feedback providers themselves. Becoming an observer and executing observations is a voluntary cooperative act: it is freely chosen, it is costly and most of the benefits are accrued by fellow co-workers that receive the feedback. As expected, the incentive to cooperate decreases as the number of observers grow, limiting the evolution of cooperation.
We designed three treatments that attempt to ameliorate the problem of cooperation in large groups. We implement these treatments in four sites:
Treatment 1. In each of the four sites, we created a structure of five groups where one tenth of the employees were randomly assigned to five randomly selected observers. The selected observers, as well as the enrolled ones, will be bound to execute their observations within the assigned group.
Treatment 2. In the sites 2 and 4, we allowed the observers in these groups to meet regularly and to interact with originals ten observers of the site.
Treatment 3. In the sites 3 and 4, we will publish a monthly list of the amount of observations executed by all the observers in the site.

Experimental design

Our experimental design is as follows:
- We execute a field experiment in four stores of a large home improvement retail company.
- In these stores, the company is implementing a methodology designed to improve workplace safety. In this method, a group of ten workers are selected (forming the "site committee") and then they are trained to: i) provide 10 minutes feedback on safety behavior to fellow workers (becoming "observers"), ii) invite and train other workers of the store to become "observers". Which worker is observed is not structured nor directed by the standard implementation of the methodology: although some degree of structure might emerge over time in an implementation, there is a lot of random observations in the process (e.g., "just observing the worker at hand").
- The key issue in this methodology is that becoming an observer and executing observations is costly, while most of the benefits are accrued by fellow co-workers that receive the feedback. In essence, they execute a cooperative act: bear a personal cost in order to benefit a third party. As expected, the problem in this methodology is cooperation in large groups. As the number of observers in a site grow, the incentives and the commitment to execute the expected number of observations goes down.
- We designed three treatments that attempt to ameliorate the problem of cooperation in large groups:
1) Treatment 1. In each of the four sites, we created a structure of five groups where one tenth of the employees were randomly assigned to one of five randomly selected observers. The selected observers, as well as the enrolled ones, will be bound to the assigned group to execute their observations.
2) Treatment 2. In the sites 2 and 4, we allowed these groups to meet and also to interact with the rest of the "site committee".
3) Treatment 3. In the sites 3 and 4, we will publish a monthly list of the amount of observations executed by all the observers in the site.
- To execute these treatment we collaborate with the retail store company and with the consulting company in charge of executing the methodology.

Design details

Randomization method
- The assignment of treatments to sites was done by randomizing the treatment across the sites with a spreadsheet. - The assignment of observers to groups is being done on the ground by the consultant executing the methodology. This randomization is being done by using a lottery box, where there are ten balls and only five indicate being selected to the treatment 1. The ten observers have to select a ball from the box. - The assignment of the employees to groups was done using a Stata randomization command. We generated four blocks (stratas) in the assignment: gender, age, tenure and position.
Randomization unit
- For assigning workers to groups, the unit was individual workers. - For assigning observers to treatment 1, the observer was the unit. When analyzing outcomes we need to consider these assignment rules. For example, we will analyze accident at the individual worker level, rendering the clustering of standard errors at the observer level a necessity. In contrast, to study observations we will compare observations between observers, generating the need to cluster standard errors within stores.
Clusters
We have four sites (four stores), and thus, four clusters.
Observations
On average, in each store we have 10 observers and, on average, 250 workers. Five observers and 125 workers will be randomly matched in groups of 25 workers (the remaining observers and workers will have a standard implementation, without group structuring). Thus, we will have 20 observers in treatment and 20 observers in control, as well as 500 workers in treatment and 500 workers in control. We will observe at least a year of observations. Each observer is supposed to execute one observation a week. We will collect monthly information on accidents and observations. We will also collect historical information on accidents (prior to the treatment) in order to improve statistical inference.
Arms
As indicated above, we will have 20 observers and 500 workers in the treatment condition and 20 observers 500 workers in the control condition.
Primary outcomes
There are two main outcome variables we are interested in: 1) Number of observations per observer. We expect that the treatments will make the observers do a higher number of observations. 2) Workplace accidents. We expect that the workers assigned to the groups will experience a lower number of accidents. 3) Percentage of workers that become observers. We expect this percentage to be higher in our treatments conditions.

Cited by

No citation data available for this registration yet.

Keywords